Toolbox Design

Our toolbox was designed to allow clinicians and researchers to quickly ask questions about pupillary response without having to write processing scripts. To this end, we developed our toolbox with a Graphic User Interface (GUI) that allows non-programmers to investigate the mean pupillary response curves for up to 3 categories of interest, define a feature selection methodology, or choose from preset feature selection techniques, and explore how well classification and regression models fit the data based on the selected features.

​

We use cross validation and grid searching to optimize and test Support Vector Machine and Ridge Regression models on user-defined data. Finally, we use the vector of values predicted by these algorithms for each image in the assay to classify a binary property of each subject, such as depression.

​

Our preprocessing module replaces missing values in input data via linear interpolation. This preprocessing module also filters out samples that don’t meet user parameters. We then use feature selection to reduce the dimensionality of the samples. Principal Component Analysis (PCA), Mean Value, Total Curve, and Receiver Operating Characteristic (ROC) plotting are called via GUI for quick visualization.